About this document
NUDGE: Interpretable RL with Logic Policies by 骥Zhang is a document available to read on EtoBox.
1) The document proposes NUDGE, a method for producing interpretable and explainable reinforcement learning agents through the use of differentiable logic. 2) NUDGE represents policies as weighted logic rules and uses differentiable forward reasoning to select actions and compute explanations using input gradients. 3) NUDGE employs neurally-guided symbolic abstraction to efficiently learn rulesets by using pretrained neural agents, producing agents that are interpretable, explainable, and robust to envi
- Author
- 骥Zhang
- Language
- EN